Diffuse optical tomography guided adaptive reconstruction in fluorescence molecular tomography

During the past decade, optical imaging combined with tomographic approaches has proved its potential in offering quantitative three-dimensional spatial maps of chromophore or fluorophore concentration in vivo. Due to its direct application in biology and biomedicine, diffuse optical tomography (DOT) and its fluorescence counterpart, fluorescence molecular tomography (FMT), have benefited from an increase in devoted research and new experimental and theoretical developments, giving rise to a new imaging modality. The most recent advances in FMT and DOT are based on the capability of collecting large data sets by using CCDs as detectors, and on the ability to include multiple projections through recently developed noncontact approaches. For these to be implemented, we have developed an imaging setup that enables three-dimensional imaging of arbitrary shapes in fluorescence or absorption mode that is appropriate for small animal imaging. This is achieved by implementing a noncontact approach both for sources and detectors and coregistering surface geometry measurements using the same CCD camera. A thresholded shadowgrammetry approach is applied to the geometry measurements to retrieve the surface mesh. We present the evaluation of the system and method in recovering three-dimensional surfaces from phantom data and live mice. The approach is used to map the measured in vivo fluorescence data onto the tissue surface by making use of the free-space propagation equations, as well as to reconstruct fluorescence concentrations inside highly scattering tissuelike phantom samples. Finally, the potential use of this setup for in vivo small animal imaging and its impact on biomedical research is discussed.

Learning-based methods have been proved to perform well in a variety of areas in the biomedical field, such as biomedical image segmentation, and histopathological image analysis. Deep learning, as the most recently presented approach of learning-based methods, has attracted more and more attention. For instance, massive researches of deep learning methods for image reconstructions of computed tomography (CT) and magnetic resonance imaging (MRI) have been reported, indicating the great potential of deep learning for inverse problems. Optical technology-related medical imaging modalities including diffuse optical tomography (DOT), fluorescence molecular tomography (FMT), bioluminescence tomography (BLT), and photoacoustic tomography (PAT) are also dramatically innovated by introducing learning-based methods, in particular deep learning methods, to obtain better reconstruction results. This review depicts the latest researches on learning-based optical tomography of DOT, FMT, BLT, and PAT. According to the most recent studies, learning-based methods applied in the field of optical tomography are categorized as kernel-based methods and deep learning methods. In this review, the former are regarded as a sort of conventional learning-based methods and the latter are subdivided into model-based methods, post-processing methods, and end-to-end methods. Algorithm as well as data acquisition strategy are discussed in this review. The evaluations of these methods are summarized to illustrate the performance of deep learning-based reconstruction.

Diffuse optical tomography (DOT) is to reconstruct the images of internal optical parameters distribution from boundary measurements. Due to the amount of available boundary measurements is less than the number of unknown optical parameters to be recovered, this inverse problem usually shows the ill-posed characteristics. This will result in the problem of low reconstruction image quality. In this paper, an adaptive regularization method based on the objective function values is proposed, which reduces the ill-posed characteristics in the inverse problem by selecting an appropriate regularization value at teach iteration. Results from computer simulations indicated that using this regularization technique, DOT imaging quality is improved effectively. Furthermore, using the regularization technique, the sensitivity to noise of the reconstructed images can be decreased greatly.© (2007) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

International audience

Optical imaging of fluorescent probes is an essential tool for investigation of molecular events in small animals for drug developments. In order to get localization and quantification information of fluorescent labels, CEA-LETI has developed efficient approaches in classical reflectance imaging as well as in diffuse optical tomographic imaging with continuous and temporal signals. This paper presents an overview of the different approaches investigated and their performances. High quality fluorescence reflectance imaging is obtained thanks to the development of an original "multiple wavelengths" system. The uniformity of the excitation light surface area is better than 15%. Combined with the use of adapted fluorescent probes, this system enables an accurate detection of pathological tissues, such as nodules, beneath the animal's observed area. Performances for the detection of ovarian nodules on a nude mouse are shown. In order to investigate deeper inside animals and get 3D localization, diffuse optical tomography systems are being developed for both slab and cylindrical geometries. For these two geometries, our reconstruction algorithms are based on analytical expression of light diffusion. Thanks to an accurate introduction of light/matter interaction process in the algorithms, high quality reconstructions of tumors in mice have been obtained. Reconstruction of lung tumors on mice are presented. By the use of temporal diffuse optical imaging, localization and quantification performances can be improved at the price of a more sophisticated acquisition system and more elaborate information processing methods. Such a system based on a pulsed laser diode and a time correlated single photon counting system has been set up. Performances of this system for localization and quantification of fluorescent probes are presented.

The accuracy and clinical utility of diffuse optical imaging systems rely on continuous advancements in instrumentation, theoretical modeling, and computational methods, alongside strategies to overcome persistent barriers to clinical translation. A key challenge in diffuse optical imaging research is the development of computationally efficient numerical models that accurately describe light-tissue interactions in complex anatomical domains, along with the validation of these models and novel instrumentation using realistic tissue-mimicking phantoms. Although diffuse optical imaging remains a promising area of research for applications such as cancer detection and functional brain activity assessment, progress has been limited by the lack of anatomically complex optical phantoms and the high computational cost of Monte Carlo simulations required to model light propagation. This dissertation addresses these challenges through two key innovations. First, we designed a multi-filament-mixing 3-D printing platform which enables spatially varying absorption, scattering, and luminescent optical properties by dynamically blending multiple commercially available filaments. This platform facilitates the design and fabrication of anatomically accurate phantoms with precise control over optical absorbance and scattering parameters, while also incorporating tunable luminescence intensity through the embedding of phosphorescent materials. We printed several benchmark phantoms showcasing our method's capability of generating complex anatomical shapes with continuously varying optical properties. This method should therefore allow for the design and easy fabrication of highly complex, anatomically accurate phantoms for benchmarking, calibration, and training across a wide range of diffuse optical imaging applications. Second, we introduce a memory-efficient parametric modeling approach to address the computational challenges in capturing detected photon partial path length (PPL) distributions in Monte Carlo simulations. This method utilizes numerical stochastic estimation techniques, such as the Box-Cox transformation, skew-normal distribution modeling, and rejection sampling, to efficiently approximate the PPL distribution with a small set of statistical parameters. This approach significantly reduces the memory footprint of large-scale Monte Carlo simulations, while remaining generalizable to anatomically accurate domains. Furthermore, the ability to regenerate an arbitrary number of detected photon PPLs using the recorded parametric PPL models demonstrates significant computational savings while retaining high levels of accuracy. Together, these findings enable more accessible and accurate diffuse optical imaging modeling by substantially reducing the computational costs typically associated with simulating light transport in complex anatomical domains. In summary, this dissertation presents experimental and computational advancements that are broadly applicable to diffuse optical imaging research. Both innovations are generalizable to complex tissue-like domains, enhancing their relevance across a wide range of applications. These contributions support the continued development of diffuse optical imaging technologies and promote their clinical translation, helping to enable more accessible, cost-effective, and practical imaging systems.--Author's abstract

We present high spatial density, multi-modal, parallel-plate Diffuse Optical Tomography (DOT) imaging systems for the purpose of breast tumor detection. One hybrid instrument provides time domain (TD) and continuous wave (CW) DOT at 64 source fiber positions. The TD diffuse optical spectroscopy with PMT- detection produces low-resolution images of absolute tissue scattering and absorption while the spatially dense array of CCD-coupled detector fibers (108 detectors) provides higher-resolution CW images of relative tissue optical properties. Reconstruction of the tissue optical properties, along with total hemoglobin concentration and tissue oxygen saturation, is performed using the TOAST software suite. Comparison of the spatially-dense DOT images and MR images allows for a robust validation of DOT against an accepted clinical modality. Additionally, the structural information from co-registered MR images is used as a spatial prior to improve the quality of the functional optical images and provide more accurate quantification of the optical and hemodynamic properties of tumors. We also present an optical-only imaging system that provides frequency domain (FD) DOT at 209 source positions with full CCD detection and incorporates optical fringe projection profilometry to determine the breast boundary. This profilometry serves as a spatial constraint, improving the quality of the DOT reconstructions while retaining the benefits of an optical-only device. We present initial images from both human subjects and phantoms to display the utility of high spatial density data and multi-modal information in DOT reconstruction with the two systems.

A region-based approach of image reconstruction using the finite element method is developed for diffuse optical tomography (DOT). The method is based on the framework of the pixel-based DOT methodology and on an assumption that different anatomical regions have their respective sets of the homogeneous optical properties distributions. With this hypothesis, the region-based DOT solution greatly improves the ill-posedness of the inverse problem by reducing the number of unknowns to be reconstructed. The experimental validation of the methodology is performed on a solid phantom employing a multi-channel DOT system of lock-in photon-counting mode, as well as compared with the traditional pixel-based reconstruction results, demonstrate that the proposed DOT methodology presents a promising tool of in vivo reconstructing background optical structures with the aid of anatomical a priori . Key words: Diffuse optical tomography, Region-based reconstruction, Pixel-based reconstruction, Optical properties.

We introduce novel adaptive methods to approximate moments of solutions of partial differential Equations (PDEs) with uncertain parametric inputs. A typical problem in Uncertainty Quantification is the approximation of the expected values of quantities of interest of the solution, which requires the efficient numerical approximation of high-dimensional integrals. We perform this task by a class of deterministic quasi-Monte Carlo integration rules derived from Polynomial lattices, that allows to control a-posteriori the integration error without querying the governing PDE and does not incur the curse of dimensionality. Based on an abstract formulation of adaptive finite element methods (AFEM) for deterministic problems, we infer convergence of the combined adaptive algorithms in the parameter and physical space. We propose a selection of examples of PDEs admissible for these algorithms. Finally, we present numerical evidence of convergence for a model diffusion PDE.

A crucial parameter in Diffuse Optical Tomography (DOT) is the construction of an accurate forward model, which greatly depends on tissue boundary. Since photon propagation is a three-dimensional volumetric problem, extraction and subsequent modeling of three-dimensional boundaries is essential. Original experimental demonstration of the feasibility of DOT to reconstruct absorbers, scatterers and fluorochromes used phantoms or tissues confined appropriately to conform to easily modeled geometries such as a slab or a cylinder. In later years several methods have been developed to model photon propagation through diffuse media with complex boundaries using numerical solutions of the diffusion or transport equation (finite elements or differences) or more recently analytical methods based on the tangent-plane method . While optical examinations performed simultaneously with anatomical imaging modalities such as MRI provide well-defined boundaries, very limited progress has been done so far in extracting full-field (360 degree) boundaries for in-vivo three-dimensional DOT stand-alone imaging. In this paper, we present a desktop multi-spectrum in-vivo 3D DOT system for small animal imaging. This system is augmented with Technest's full-field 3D cameras. The built system has the capability of acquiring 3D object surface profiles in real time and registering 3D boundary with diffuse tomography. Extensive experiments are performed on phantoms and small animals by our collaborators at the Center for Molecular Imaging Research (CMIR) at Massachusetts General Hospital (MGH) and Harvard Medical School. Data has shown successful reconstructed DOT data with improved accuracy.

Diffuse optical tomography (DOT) is a non-invasive imaging modality that uses near-infrared light to probe the optical properties of tissue. In conventionally used deterministic methods for DOT inversion, the measurement errors were not taken into account, resulting in unsatisfactory noise robustness and, consequently, affecting the DOT image reconstruction quality. In order to overcome this defect, an extended Kalman filter (EKF)-based DOT reconstruction algorithm was introduced first, which improved the reconstruction results by incorporating a priori information and measurement errors to the model. Further, to mitigate the instability caused by the ill-condition of the observation matrix in the tomographic imaging problem, a new, to the best of our knowledge, estimation algorithm was derived by incorporating Tikhonov regularization to the EKF method. To verify the effectiveness of the EKF algorithm and Tikhonov regularization-based EKF algorithm for DOT imaging, a series of numerical simulations and phantom experiments were conducted, and the experimental results were quantitatively evaluated and compared with two conventionally used deterministic methods involving the algebraic reconstruction technique and Levenberg-Marquardt algorithm. The results show that the two EKF-based algorithms can accurately estimate the location and size of the target, and the imaging accuracy and noise robustness are obviously improved. Furthermore, the Tikhonov regularization-based EKF obtained optimal parameter estimations, especially under the circumstance of low absorption contrast (1.2) and high noise level (10%).

Diffuse optical tomography (DOT) is a noninvasive biomedical imaging method to reconstruct optical property distribution. Since the underdetermined characteristic of reconstruction process, a priori information such as the structure provided by multimodal images are beneficial for imaging quality. We introduce a deep convolutional neural network-based method to rapidly calculate the heterogenous region by the diffusive intensity distribution measured by the same device used for DOT imaging. The process is based on a convolutional forward model which can accurately calculate the diffusive light intensity distribution with known structure and corresponding optical properties. The heterogeneous region imaging network is the inverse of the forward model and trained with Monte Carlo simulation results. The trained inverse network achieves the imaging sensitivity and specificity of 0.91 and 0.89 for validation data-set and the reconstruction speed is under 0.1s peer image.

Near infrared spectroscopy (NIRS) and diffuse optical tomography (DOT) of the brain reveal no information about the measurement's underlying anatomical structures. An independent anatomical mapping of DOT results onto the subject's brain or a generic brain model is desirable, especially when regions prone to large inter-subject variability are studied. We show two methods to match DOT data from high density fiber grids to anatomical structures. The forward model that is used to predict the light propagation is based on one generic anatomical MR scan. In both approaches we use this model MR-scan to translocate the position of the optical fiber grid from our experimental setup to the FEM model space. The first method, using fiduciary marks, achieves the spatial normalization of the subject's MR-scan (with marked corners of the fiber grid) and the model's MR scan, leading to a translocation of the fiber pad position to the FEM-Model space. The second, anatomic landmark based, approach does not require the individual's MR scan. For this, 19 reference points and the position of the fiber pad corners are determined using photogrammetry software. These coordinates are translocated to the FEM model space by solving the least square problem of the subject's and the model's reference points. We illustrate and compare both methods and show results from a vibrotactile stimulation experiment in humans.

There are different studies that allow detecting the presence of cancer cells in a patient; however, the time it takes to obtain a correct diagnosis is critical for these cases. This work presents the design and construction of a prototype as first approach for a microtomograph based on Diffuse Optical Tomography (DOT). Diffused Optical Tomography is a technique that uses light from the Near Infrared Region (NIR) to measure tissue’s optical properties that has become relevant for the medical field due to it being a non-invasive technique. The main goal of this device is to be able to detect and analyze cancerous tumors by measuring diffuse photon densities in turbid media. As a first phase of the development, this project integrates an image acquisition mechanism, an image processing interface at software level developed with Blender to exploit a GPU architecture to optimize the execution time.

Anthropomorphic Polydimethylsiloxane silicone-based phantom for Diffuse Optical Imaging